oldworld benchmarks & X funny business
[If you don't care about OldWorld Macs, you can safely hit DELETE now.]
All the software and packages mentioned in this post are from the stable release.
So I got this 256KB cache module for the Ol' 7200/75, which has a hard time keeping up with the last 486 I used, and it seemed to run faster. But I thought I should have some scientific data to really tell, so I ran the following exhaustive benchmark to see what I could see.
Compile the latest official stable kernel source [2.2.18pre21] from debian with the powerpc patch applied. X running, with a couple of system monitors running on the logged in X console.
make vmlinux w/o cache: 60 minutes
make vmlinux w/256KB cache: 45 minutes.
Hmm, looks like a clear winner of about 25%.
Curiouser still, I have an x86 which is my main workstation (it's actually a laptop), and as such I usually run it with the kupdate flushing daemon disabled. But one thing I have noticed is that I can compile the entire kernel without so much as a single disk access when in this mode [type "sync" afterwards and take a coffee break], and BOY do kernel compiles go a lot faster than normal. "Normal" being about 3 minutes, but in this being less than 2.5 minutes. Also, I use a special patched gcc with mega optimization turned on. Anyway I thought I would try this on the 7200.
With several X server crashes, it took 48 minutes. Probably because of the X server crashes and because it did access the disk many times. The disk accesses were probably because it only has 40MB main mem and the laptop has 256MB.
OK, so back to the X server crashes.
These happen to me with some regularity on my two OW macs with the stock stable 2.2.18pre21 kernel and the stock X server. Not all that much when actually using X on the console, but almost with clock like regularity if I switch to a virtual console, just a few seconds after switching, X will crash, and wdm will restart it. Not that I care that much, as X on a 640x480 screen is not too useful. [Anybody have a clue how to get it to do 800x600?] But still I thought it was sorta quaint. But in the case of the kernel compile benchmark, it was crashing every ten minutes or so, and I wasn't anywhere near the console.
That was when I noticed that I was running one of these freshly compiled kernels of my own configging, and also that I had config'd it with support for only the `control' video controller, not with valkerie or platinum. But /proc/fb and the X server log said that this was a platinum controller (who knew?), so I recompiled with just platinum, but neither control or valkerie, and I haven't had a X server crash since. Interesting, no?
Also, I'm running WindowMaker in 16 bit mode and it works just fine for me. No problems, and the colors look fine. Not good or great, on this ancient 5" monitor, OK, it's really something like 14", but dang it seems small.
These things get funner every day.